evidence on the determinants of the choice …web.stanford.edu › ~rehall › hk051310.pdfevidence...
Post on 27-Jun-2020
5 Views
Preview:
TRANSCRIPT
Evidence on the Determinants of the Choice between Wage Posting
and Wage Bargaining ∗
Robert E. HallHoover Institution and Department of Economics,
Stanford University
National Bureau of Economic Research
rehall@stanford.edu
Alan B. KruegerUnited States Treasury
On leave from the Woodrow Wilson School and Department of Economics,
Princeton University
akrueger@princeton.edu
May 13, 2010
AbstractSome workers bargain with prospective employers before accepting a job. Others
face a posted wage as a take-it-or-leave-it opportunity. Theories of wage formationpoint to substantial differences in labor-market equilibrium between bargained andposted wages. We surveyed a representative sample of U.S. workers to inquire aboutthe wage determination process at the time they were hired into their current or mostrecent jobs. A third of the respondents reported bargaining over pay before acceptingtheir current jobs. About a third of workers had precise information about pay whenthey first met with their employers, a sign of wage posting. About 40 percent of workerscould have remained on their earlier jobs at the time they accepted their current jobs,indicating a more favorable bargaining position than is held by unemployed job-seekers.Our analysis of the distribution of wages shows that wage dispersion is higher amongworkers who bargained for their wages. Wages are higher among bargainers than non-bargainers, after adjusting for the differing compositions of the groups. Our results onwages give substantial support to the job-ladder model—workers who had the optionto remain at their earlier jobs when they took their current jobs earn higher wagesthan those without that option.
∗Hall’s research is part of the program on Economic Fluctuations and Growth of the NBER. The opinionsstated in this paper are those of the authors alone and not the U.S. Treasury. We are grateful to our discussantJohn Kennan, to James Malcomson, Guido Menzio, and Giuseppe Moscarini, and to numerous conferenceparticipants for helpful comments. A file containing the calculations is available by googling the first author.
1
1 Introduction
Labor is one of the most heterogeneous products traded in a modern economy. The compet-
itive market for a commodity, where all units are interchangeable and all trade for the same
price, could hardly be a worse description of the labor market. No Walrasian auctioneer
determines the wage. We study survey evidence on the ways that an employer and a worker
determine the wage at the outset of their relationship. Our findings support the predictions
of theories of wage determination about the relationship among the level and dispersion of
wages, on the one hand, and the incidence of bargaining, on the other hand
The extensive literature on this topic considers two main cases. The first is wage posting.
Here an employer defines a job in terms of duties and qualifications, and commits to a
wage. If a candidate is found qualified and interested, the employer offers the wage on a
take-it-or-leave-it basis. The second is bargaining. The employer makes an initial offer, but
the candidate can make a counteroffer for a higher wage, if so inclined. A key difference
between the two modes is the employer’s commitment not to entertain a counteroffer. To
the employer, the advantage of posted wages is that the employer may appropriate a large
fraction of the surplus of a match. The disadvantage is that a posted wage precludes a
match with a candidate whose reservation wage is higher than the posted wage but whose
productivity is even higher. Bargaining with this worker would have gained some of the
surplus. Bargaining is in the interest of the employer if qualified workers have heterogeneous
skill levels.
In the United States, a small fraction of workers in private employment and a larger
fraction in government employment receive pay under the terms of collective bargaining
agreements. From the point of view of an individual worker, the resulting wage is posted
rather than bargained individually. We identify unionized and government workers in our
analysis of the survey data.
In addition to wage posting and bargaining, one could imagine a labor market where
employers can commit to ignore counteroffers, on the one hand, but make a custom offer
to the applicant rather than offering the same wage to all qualified applicants, on the other
hand. This market would encounter the Diamond paradox. An employer would make an
offer that just meets a worker’s reservation wage. The worker knows that the wage is below
the maximum that would be acceptable to the employer, but also knows that the employer
will not consider a counteroffer that is below that maximum but better than the employer’s
2
original offer. The worker will accept the original offer. The only equilibrium in the labor
market under these conditions is for workers to earn the bare minimum needed to attract
them to the market. Workers would not earn the Ricardian rents that normally make up a
substantial fraction of wages. We do not believe that this equilibrium occurs in U.S. labor
markets. Most workers, especially men aged 25 to 55, appear to have strong preferences for
higher market goods consumption together with the higher market work needed to finance
that consumption rather than lower levels of both, a clear sign that they have not been
pushed to their participation reservation points.
Wage formation has a central role in the theory of unemployment. A positive level
of unemployment is inevitable given the frictions in the labor market—some workers will
always be in the process of locating a better use for their services after a decline in their
productivity in earlier jobs. The anticipated wage determines the payoff to workers to search
for new jobs and to employers to recruit new workers. Some bargaining protocols imply that
the bargained wage is insensitive to conditions in the labor market, such as productivity and
unemployment—see Hall and Milgrom (2008). In these models, the incentive for employers
to recruit new workers falls in times of low productivity. As a result, unemployment is
sensitive to driving forces such as productivity; the models can deliver realistic volatility of
unemployment. By contrast, in models where employers post wages and adjust them each
period to their optimal levels, the response of unemployment to driving forces tends to be
small and the observed volatility of unemployment remains unexplained.
Our survey has about 1,400 respondents who took jobs sufficiently recently that we believe
that their answers about wage formation at the beginning of the job were reasonably reliable.
In addition to many questions about their backgounds, we asked four questions that bear
specifically on wage formation. The first determined how much a respondent knew about
pay before being interviewed for the job. This question bears on the public nature of the
wage in a wage-posting market. The second asked if the wage offer for the current job was
take-it-or-leave-it or if bargaining occurred. The third asked if the respondent could have
kept an existing job at the time he or she took the current job. The option to keep an
existing job is valuable in a bargaining setting. The fourth asked if the employer learned the
respondent’s earlier pay rate during the evaluation process. This knowledge would improve
the employer’s expected benefit in a setting with bargaining.
Our results show that both major models of wage formation have important roles in the
3
U.S. labor market. We find a fairly high level of knowledge among job-seekers prior to their
job interviews, potentially the result, in part, of public wage posting. We confirm that this
information is particularly common among union members and those who took government
jobs. We document a sharply negative relation between education and precise information
about pay—non high-school graduates are almost twice as likely as those with professional
education to know prospective pay exactly. Thus wage posting appears to be much more
important in the less heterogeneous jobs available to those with less education. We discuss
the literature that supports the proposition that wage posting is more likely to appear in
equilibrium in a market where workers have more heterogeneous skills within groups defined
by observed and verifiable characteristics.
We find that about a third of all workers bargained with their current employers—they
did not consider their job offers to be take-it-or-leave-it. Bargaining is more common by
minority workers and less common by women. The education gradient for bargaining is
remarkably steep, rising from 28 percent for those who did not graduate from high school
to 56 percent for those with professional degrees. Individual bargaining is rare for union or
government jobs.
The respondents who knew the wage in advance and who understood the wage offer to
be take-it-or-leave-it are the most likely to have taken jobs in a market with posted wages.
We find that this group is 15 percent of workers if we interpret knowing as knowing exactly
the wage when first interviewed by the employer. We find that minorities are somewhat
less likely to be in the posted-wage sector and women more likely. Participation in the
posted-wage sector falls strongly with education. Union members and government workers
are highly likely to hold posted-wage jobs.
We find that about 48 percent of workers could have kept their earlier jobs at the time
they were considering their current jobs. A substantial fraction of these workers bargained
for the wage on their current jobs; virtually all those in a group we call knowledge workers
bargained. We conclude that job-ladder models emphasizing on-the-job search are highly
relevant for understanding wage formation and that, for jobs with wage bargaining, the
option to stay in the earlier job should have an important influence on the wage bargain.
We find that those holding the option earn $2.07 more per hour than those who did not have
existing jobs to fall back on when they took their current jobs.
Finally, we find that 48 percent of workers reported that their employers had learned their
4
pay in their earlier jobs before making the offer that led to the current job. This fraction
is slightly higher for jobs where actual wage bargaining occurred. It is about the same in
situations where the respondent could have kept an earlier job. The fraction varies only a
little among categories of workers.
Our survey contains the standard questions from the Current Population Survey about
wage rates. We investigate two aspects of the relation between bargaining and the dis-
tribution of wages, conditional on observed characteristics such as education. The first is
dispersion. If the distinction between wage posting and bargaining is meaningful, the dis-
persion of wages among workers who accepted posted wages should be smaller than the
dispersion among those who bargained. This prediction holds no matter what factors result
in the choices of employers about the mode of wage determination, but it is all the stronger if
bargaining is more common for jobs that attract heterogeneous applicants. The second issue
is the level of wages of observationally similar workers whose wages were posted or bargained.
Theory suggests that wages resulting from bargaining may be higher than posted wages, on
the average, a suggestion that our survey supports.
2 Theory
2.1 Research on wage formation
Pissarides (2000), Mortensen (2003), and Rogerson, Shimer and Wright (2005) discuss much
of the relevant research on wage formation for individual workers. In the brief summary in
this section, we will discuss all of the research as if it had originally referred to the labor
market, though a number of important papers were actually stated in terms of product
markets.
Stigler (1961, 1962) launched modern thinking about trade in markets for heterogeneous
products. He observed that, in the presence of variation in wages, a job-seeker should sample
from the distribution of available wages. The choice of the number of samples balances the
benefit of finding a higher wage, where the marginal benefit falls with sample size, against
the cost of sampling, which is assumed to be constant. McCall (1970) provided an analytical
solution to this problem, based on class notes from Kenneth Arrow. Unemployment depends
on the amount of sampling job-seekers find optimal.
Diamond (1971) tackled the problem of equilibrium in a search market. He considered
the optimal policy of an employer confronted by searchers who follow the Stigler-McCall
5
prescription. He observed that, if all other employers set the same wage, one employer
could still hire a random visitor by setting a wage below the common wage, but not so
far below as to cause the visitor to incur the cost of visiting another employer. This logic
shows that a common wage cannot be an equilibrium in the market, with one exception: the
monopsony wage could be an equilibrium, because the employer, though able to get a lower
wage, would choose not to, as a wage below the monopsony level would yield a lower profit.
Diamond concluded that the only common wage in equilibrium would be the monopsony
wage, a proposition known as the Diamond paradox. In a simple model where hours of
work for those choosing to participate in the labor market are fixed at a standard level or
are fairly inelastically supplied, the monopsony wage is the reservation wage that is just
sufficient to induce workers to participate in the labor market. This wage level is the value
of some alternative activity available to the job-seeker. If the wages for all jobs for which a
worker is qualified are set in the way Diamond contemplated, the only equilibrium common
wage leaves the worker indifferent between work and non-work. That is, the common wage
deprives the worker of all of the Ricardian rents that we normally associate with the labor
market, where the elasticity of supply for most workers is relatively low and the area above
the supply curve describing the rents is correspondingly large. The supply curve is perfectly
elastic at the indifference point, so the area is zero in the Diamond paradox.
The Diamond paradox rests on a strong assumption: The hapless job-seeker cannot strike
back by making a counteroffer. The same logic that permits the employer to make a low
offer that is still acceptable applies equally to the job-seeker, whose higher counteroffer could
leave the employer better off than not hiring the job-seeker. In the Diamond paradox, the
employer has all of the bargaining power. Somehow the employer has the power to disregard
a counteroffer. It is an open question whether disregarding a counteroffer is credible. The
general standard for credibility is on-the-spot rationality or subgame perfection. There is
no general answer as a matter of theory to the question of whether an employer holds all
the bargaining power. We regard the wage-formation process that leads to the Diamond
paradox as unrealistic, not because we know that it is impossible for an employer to have
all the bargaining power, but because we believe that wages are well above the indifference
level for many workers.
Butters (1977) responded to Diamond’s challenge in a way directly relevant to this paper:
In Butters’s model, the employer offers the same wage to all job-seekers, but the job-seekers
6
differ among themselves with respect to information about other wages, so they have diverse
reservation wages. By failing to customize the offer, the employer gives some of the surplus
to all but the applicant with a reservation wage equal to the common value of all of the
employer’s offers. The Diamond paradox arises from a situation where the wage offered to
each job-seeker is at the reservation level of that person. Butters launched the posted-wage
model that we investigate here. In his model, job-seekers learn randomly about the wages
of some but not all jobs. In equilibrium, the distribution of wages does not collapse to
a single value. Burdett and Judd (1983) extended Butters’s and other formulations and
discussed other models that portray heterogeneous reservations wages but a common offered
wage. Burdett and Mortensen (1998) developed a complete equilibrium model of the labor
market along these lines in which the needed heterogeneity arises from job search by employed
workers. Because they retain the option of keeping their current jobs, their reservation wages
are equal to their current wages and are above the common reservation wage of unemployed
job-seekers. Burdett and Mortensen’s paper demonstrated the importance of on-the-job
search as a matter of theory. Quantification of its importance has proven difficult, as it
rests on the fraction of workers who retain the option to stay on their current jobs as they
investigate a possible new job. On-the-job search is easy to define in a model but hard to
measure in practice.
Within posted-wage models, there is an interesting question as to the public information
about wages. Models of directed search give job-seekers partial or complete information
about the terms of employment. The simplest setup of this type, with full information
available for free to all job-seekers about all job openings, collapses to perfect competition.
Rogerson et al. (2005) discuss models where frictions remain.
The other main branch of individual wage-formation theory attributes more symmet-
ric roles to the job-seeker and employer by assuming that they bargain with each other.
Mortensen (1978) and Diamond and Maskin (1979) began this line of thought in the context
of the formation of symmetric pairs. Diamond (1982) applied it to the labor market with
job-seeker-employer matching and wage formation. The modern canon is Mortensen and
Pissarides (1994). This literature generally adopted the Nash wage bargain.
Nash’s development of his eponymous bargain does not tell a story about how the bar-
gainers reach their bargain. Alternating-offer bargaining seems to fit the way that bargains
are actually made in the labor market and elsewhere. Hall and Milgrom (2008) apply Bin-
7
more, Rubinstein and Wolinsky’s (1986) version of the alternating-offer bargaining model to
the setting of the Mortensen-Pissarides model (the paper also discusses earlier applications
of alternating-offer bargaining in the labor market). The convention of the labor market
appears to be that the employer makes the first offer, just as the seller in the housing market
makes the first offer in the form of an asking price. But the wage offer is not a posted
wage—the applicant is free to respond with a counteroffer, just as the potential buyer may
offer to pay less than the asking price for a house. The parties continue to exchange offers
until the prospective benefit from continuing falls short of the cost of delaying the deal, at
which point one side accepts the other’s offer and the deal is done.
A key point is that the unique equilibrium of an alternating-offer bargaining process
with full information is for the employer to offer an acceptable wage as the first offer—one
just at the margin between triggering a counteroffer and not triggering one. Thus with
full information, one never observes parties actually making counteroffers. One could easily
confuse a posted wage setup with an alternating-offer setup, because in both cases, the
employer would say, “The job pays X. Do you accept it?” and the job-seeker would just say
yes or no, without a counteroffer.
The Nash bargain gives primacy to the bargainers’ outside options. Within the alternating-
offer framework, the outside options are controlling if there is a reasonable probability that
an interruption would occur in the exchange of offers, if such an exchange occurred hypo-
thetically (as noted above, an actual exchange is off the equilibrium path). The interruption
could take the form of the disappearance of the employment opportunity, the appearance of
another employer willing to hire the worker, or the appearance of another worker available
to fill the job. Hall and Milgrom (2008) discuss a major difference between alternating-offer
bargaining with a low probability of interruption and bargaining with a higher probabil-
ity (Nash). With Nash, the worker’s alternative to accepting the employer’s current offer
is to re-enter the search process to find another job opening (because the opening under
consideration may disappear) or to accept a job with another employer. The value of this
alternative depends on the tightness of the labor market—the job-seeker’s bargaining po-
sition is stronger if the next job prospect is easy to find or if the likelihood of receiving a
second offer is higher. By contrast, with alternating-offer bargaining and little likelihood
of interruption, the alternative is to delay the formation of the job match. The value of
this alternative is not diminished when unemployment is higher. Thus the wage is not as
8
sensitive to unemployment. The paper explains why the lower sensitivity helps explain the
large increases in unemployment that occur in recessions.
With imperfect information, bargaining models become more complicated and prone to
indeterminacy. Hall and Lazear (1984) lay out some of the issues in the setting of individual
wage formation. See Ausubel, Cramton and Deneckere (2002) for a survey and Menzio (2007)
for an application with individual wage formation. Our survey shows that many job-seekers
actively bargain with employers, contrary to the full-information bargaining model, so this
is an appropriate area for further development.
Figure 1 summarizes the types of model of wage formation between individuals and
employers. At the top we put the issue that divides wage posting from bargaining—can
the employer commit not to respond to a counteroffer from a job-seeker or otherwise avoid
bargaining? If so, the second issue is whether the employer can make a customized offer to
the job-seeker that captures all of the surplus from the match. If so, and if the job-seeker
has no opportunities in labor markets where workers share the surplus, workers will be paid
only enough to bring them into the labor market and will enjoy no Ricardian rents. This
condition is the Diamond paradox. If wage offers are not customized and a force such as on-
the-job search causes reasonable heterogeneity in reservation wages among applicants who
will be paid the same wage, the employer’s choice of that wage will result in a sharing of the
surplus by all but those with a reservation wage equal to the wage paid.
The right side of the figure describes wage determination when the employer lacks the
ability to disregard counteroffers. In that case, alternating-offer bargaining seems the best
way to model the process. If interruptions are likely, the process resembles the Nash bargain,
where the outside option is influential. Because the current conditions in the labor market
control the value of the outside option, wages are flexible in this case. On the other hand,
if the parties perceive that interruptions are unlikely, the costs and benefits of delay in
completing the bargaining process control the outcome. Because conditions in the labor
market are less influential for these costs and benefits, the resulting wage is less connected
to those conditions. The result is a sticky wage and more volatile unemployment.
9
Commitment to ignore counteroffers
Wage offer customized to worker
Interruption to alternating offer bargaining likely?
Commitment to ignore counteroffers
Wage offer customized to worker
Diamond paradox
Posted wage
Wage tightly linked to conditions
Wage less responsive
to conditions
Interruption to alternating offer bargaining likely?
Figure 1: Models of Wage Formation
2.2 How Would an Employer Decide between Wage Posting andBargaining?
We regard the issue of the empirical importance of wage posting—that is, the employer’s
commitment not to consider counteroffers—as an open question. Commitment may fail
the test of on-the-spot rationality, a test often applied in many areas of modern economic
analysis. Employers face an issue of time consistency. They might benefit from job-seekers’
belief that counteroffers fall on deaf ears, but still occasionally encounter situations where
they make no match based on the posted wage, but would gain some of the surplus from
the match if they bargained with an applicant. On net, they would choose to sacrifice the
benefit of bargaining to gain more of the surplus in the matches they make with wage posting,
but cannot stop themselves from giving in to the temptation to consider a counteroffer in
the occasional situation where they would gain from it. In some settings of this type,
commitment via a trigger strategy may arise. The result is a reputational equilibrium—
the employer finds it advantageous to disregard a counteroffer because responding to one
would sacrifice the employer’s reputation for not considering counteroffers. The reputational
equilibrium rests on the notion that job-seekers use trigger strategies—they would switch to
making counteroffers to an employer who considered a counteroffer from a single applicant.
An alternative and perhaps more realistic way that employers can commit to disregard
counteroffers is to deny managers the power to alter the wage. Such a policy mirrors the
10
way that retailers enforce posted prices—the customer cannot find anybody in the store who
has the power to accept a price below the posted price.
2.3 Models of the choice of wage-setting mode
The most basic model of the employer’s choice between posting and bargaining takes the
only source of heterogeneity to be dispersion in candidates’ reservation wages. Suppose that
the distribution of the reservation wage, z, is
R(z) = zψ for z ∈ [0, 1]. (1)
Job-seekers encounter employers randomly, so the parameter ψ is a job-seeker’s wage elas-
ticity of labor supply to employers in general and to a particular employer. The profit per
job applicant from a strategy of posting a wage w is
(p− w)R(w). (2)
Here p is the worker’s productivity. The first-order condition for maximum profit is
w =ψ
ψ + 1p. (3)
The quantity ψψ+1
is the wage mark-down coefficient, analogous to the constant markup
coefficient for a firm facing constant elasticity of demand. Profit per applicant under the
posting strategy is
πP =ψψ
(ψ + 1)ψ+1pψ+1. (4)
The profit from a bargaining strategy is a fraction β, the employer’s bargaining power,
of the joint surplus from a match, which is the difference between productivity p and the
worker’s opportunity value z. Expected profit is
βR(p)(p− E(z|z ≤ p)). (5)
Here R(p) is the probability that the employer will make a deal with the applicant, the
probability that an efficient match is possible, with z ≤ p. The expectation of z conditional
on an efficient match isψ
ψ + 1p. (6)
Thus expected profit per applicant from a bargaining strategy is
πB =β
ψ + 1pψ+1. (7)
11
Bargaining is the more profitable strategy if
β >
(ψ
ψ + 1
)ψ
. (8)
For the benchmark case of unit-elastic labor supply, ψ = 1, the criterion is that the employer’s
bargaining power exceed one half. The critical value of β approaches 1 as the elasticity
approaches zero and approaches e−1 = 0.37 as the elasticity becomes large.
The variance of the reservation wage, z, is a hump-shaped function of the elasticity, ψ,
reaching a maximum at ψ = 0.62 (for low ψ, the density concentrates at z = 0, with close to
zero variance; for high ψ, at 1, again with small variance). Over the range from ψ = 0 to 0.62,
dispersion of the reservation wage rises with ψ and the likelihood of bargaining rises as well,
in the sense that the critical bargaining power β falls. In this range, greater heterogeneity
among job applicants inclines employers further toward bargaining rather than posting.
But the relationship reverses for higher elasticities. The assumption that the distribution
of reservation wages differs among groups only in its elasticity and not in its position is
essentially arbitrary, however. If the distribution takes the form
R(z) =
(z
γ
)ψ
for z ∈ [0, γ], (9)
and the position parameter γ is higher among groups with higher elasticity, the relation
between the variance and the critical value of bargaining power would be negative over a
wider range of values.
Under posting, all workers who accept the posted wage earn that wage, w = ψψ+1
p, so
dispersion is zero. A worker with reservation wage z ≤ p who takes a job with a bargaining
employer earns a wage of
βz + (1− β)p. (10)
The expected wage, conditional on employment, is
βψ
ψ + 1p+ (1− β)p, (11)
which is unambiguously greater than the posted wage. In a comparison of wage levels
between workers who accepted jobs with posted wages and those who bargained for their
wages, we would expect to find higher average wages in the second group. Again, this
statement is based on the essentially arbitrary assumption that the only difference between
12
the two groups was in the elasticity of the wage distribution. If the position parameter γ is
positively associated with the elasticity, then the effect would be even larger.
Our survey measures the dispersion of actual wages. Thus we expect to find lower wage
dispersion among workers who accepted posted wages than among those who bargained,
quite apart from any differences in dispersion of reservation wages or other unobserved vari-
ables. Our empirical investigation of wage dispersion in relation to posting and bargaining
is intended more to test the validity of our survey question than to provide evidence about
dispersion in those unobserved characteristics.
Ellingsen and Rosen (2003) investigate dispersion in skills rather than reservation wages
within observationally identical workers. They build a model along the lines of Mortensen
and Pissarides (1994) in which employers decide prior to meeting job-seekers whether to stick
to a posted wage or to bargain. They do not mention the obstacles to enforcing a policy
of not bargaining; they take its credibility for granted and model the choice in terms of the
maximal payoff to the employer. Heterogeneity in applicants is central to the analysis—if
all workers had the same skills and same non-market value of time, the employer would
be choosing between Nash bargaining with given division of the surplus against making a
take-it-or-leave-it offer to the applicant, which is simply Nash bargaining with all bargaining
power on the employer’s side. Obviously the employer chooses the posted wage. The authors
show that heterogeneity above a critical level makes it possible for bargaining to occur in an
equilibrium. In a bargaining equilibrium, the advantage to the employer of customizing the
wage through negotiation outweighs the reduction in the bargaining power of the employer—
capturing some of the surplus in all efficient matches yields a higher return to the employer
than capturing all of the surplus in a fraction of the efficient matches. Their model does not
rule out other equilibriums with wage posting. In particular, because they assume that the
non-market alternative for all workers has the same value, the Diamond equilibrium, where
all employers post that value as the offered wage, is an equilibrium. Michelacci and Suarez
(2006) study a directed search setup where job-seekers have a role in the choice between
wage posting and bargaining.
These two papers confirm the intuition that heterogeneity in skills may have a role in the
choice between posting and bargaining. The heterogeneity that matters is within categories
defined by verifiable characteristics of job-seekers—employers can post wages conditional on
anything verifiable. Michelacci and Suarez discuss this point.
13
2.4 Why the choice between posting and bargaining matters
The amount of rent that employers earn from the employment relationship has a central
role in models of unemployment fluctuations. Most models assign the employer the task of
deploying resources to generate job matches, which seems realistic, though we are not aware
of any systematic evidence on the subject. Matchmakers in the labor market, both on and
off the Internet, typically impose the great bulk of their charges on the employer.
The rent that employers earn in the employment relationship arises from frictions in the
labor market. When setting a wage for a job-seeker who has overcome the friction and found
a willing employer, the employer has an advantage because the job-seeker will anticipate
encountering similar friction in finding the next opportunity if the job-seeker rejects the one
in hand. Absent the friction, the job-seeker could work one employer against another until
the wage rose to the level of productivity. Because the employer gains from forming the
relationship, the employer has an incentive to deploy resources to attract job-seekers.
Variations in the rents that employers earn generate corresponding variations in the
recruiting efforts of employers. When rent is high, employers expand vacancy posting and
other recruiting activities and thus tighten the labor market, lowering unemployment. The
rent is the present value of the difference between a newly hired worker’s productivity and
the wage paid to the worker. If productivity rises and the wage remains constant or rises
less than productivity, the rent increases and the labor market tightens. If the wage tracks
productivity, labor-market tightness and unemployment remain constant.
Models of posted wages generally make the assumption of directed search—job-seekers
learn the posted wages of all employers, along with other terms of employment, and pick
the best available job. Rogerson and Shimer (2012, forthcoming), a handbook article on
job-search economics, describes this research:
A number of papers assume that firms post wages offers and workers can direct
their search towards their preferred offer, as in the competitive search literature
[citations omitted]. The equilibrium of that model coincides with the social
planners solution, and so wages are flexible. Models of wage rigidities therefore
typically assume either that firms cannot commit to wages or that workers cannot
direct their search. (footnote 31, page 41)
Although the assumption of a posted wage in a model with random search would not be
14
unreasonable, apart from the commitment issue, that combination has not appeared in the
literature, to our knowledge. Random search is paired with bargaining.
In posted-wage models where employers reset their posted wages each period, the wages
are flexible and respond fully to driving forces such as changes in productivity. Equation
(3) shows that, in the constant-elastic case, the posted wage is proportional to productivity.
Equation (11) shows that, in the same case, the expected wage resulting from bargaining
is also proportional to productivity, provided the bargaining power of the employer, β, is
constant. As a result, labor-market models with Nash bargaining generally deliver little or
no volatility of unemployment. Shimer (2010) discusses this point in detail, including the
exceptions to the general rule.
In bargaining models where the employer’s share is an endogenous variable rather than
a constant, unemployment volatility can be substantial. The direction of the movement of
the share may appear counter-intuitive—the employer’s share needs to rise in good times
with low unemployment and fall in bad times with high unemployment. To put it differently,
wages need to be sticky and respond less than fully to driving forces such as productivity.
Hall (2005) discusses this point without introducing a specific bargaining protocol, while Hall
and Milgrom (2008) show that alternating-offer bargaining with a low but realistic frequency
of interruption delivers the needed endogenous shifts of the employer’s share of the surplus.
We conclude that a finding that a substantial majority of jobs were filled at posted wages
would be unfavorable for an important branch of modern thinking about unemployment
volatility.
The New Keynesian branch of modern thinking, by contrast, presumes posted wages,
but does not characterize employers’ wage-setting as maximizing the expected benefit from
current hiring, along the lines of the simple model we presented earlier in this section. The
canon of New Keynesian economics, Christiano, Eichenbaum and Evans (2005), makes the
assumption that workers post wages. Further, a worker does not set the posted wage to its
optimal level each period, but only at random intervals, with a probability of re-optimization
of 36 percent per calendar quarter. During spells without re-optimization, the worker supplies
the amount of work that the employer chooses, given the predetermined wage. The model
does not describe unemployment explicitly, but achieves realistic volatility of total hours of
work because of the time it takes for a significant fraction of workers to adjust their wages
to a change in demand. Prior to full adjustment, employment (hours of work) is sensitive to
15
demand because of the imperfect adjustment of the wage in the shorter run.
Our survey did not inquire if workers thought they could unilaterally set their own wages.
We doubt that any of our respondents would be that presumptive. Christiano and co-authors
use the assumption for convenience, not reality. The key assumptions of the New Keynesian
model are that the wage is fixed for a significant period and that during the period, the
employer chooses the level of labor input. Whether the worker or the employer sets the wage
when it is re-optimized affects the level of the wage but not the volatility of employment
induced by persistent wage stickiness. Thus our finding that an important fraction of wages
are posted gives some support to the posting element of the New Keynesian model, but does
not bear on the question of persistent stickiness of the wage or on the idea that the employer
chooses labor input unilaterally.
3 Objectives of the Survey
We designed our survey to provide evidence on the facts underpinning the two leading types
of models of individual wage formation. The most obvious area of investigation is whether
employers post wages which, to the individual, are take-it-or-leave-it offers, as opposed to en-
gaging in any type of bargaining. We address this directly by asking if respondents perceived
the wage as take-it-or-leave-it, on the one hand, or if they bargained, on the other hand. In
formulating this question we were quite aware that there was an omitted intermediate case,
that, while the offer was not take-it-or-leave-it, the respondent did not believe that it would
be worth the effort to make a counteroffer. We regarded this as beyond the reach of our
survey techniques, though this issue is ripe for reconsideration in future work.
A second objective is to probe for information known in advance about the wage a job
paid. In the directed search model, respondents would have known the wage before sinking
any cost in applying for a job. In addition, we regard advance knowledge of the wage as
evidence of wage posting. We used two standards for assessing this knowledge: we asked
if the respondent knew the pay exactly and if the respondent had a pretty good idea. A
large fraction of the respondents picked the “pretty good idea” answer, so future work might
benefit from an intermediate phrasing.
A third objective is to measure the incidence of on-the-job search. Previous work, such
as Moscarini and Postel-Vinay (2008), has tried to answer this question from data on job
flows, but the evidence from that source is at best indirect because it does not report on the
16
issue of the option to keep the current job that is so central in wage formation theory. Also,
earlier work measures the fraction of hires that came directly from other jobs while in this
paper we measure the fraction of workers with tenure less than 10 years who retained the
option to remain on an earlier job when they were hired.
A fourth objective is to study the information that employers have about a prospect’s
current or recent wage. This information has a key role in bargaining, especially when the
prospect has the option of keeping the job rather than taking a new job with the employer.
Of course, in a survey of workers, we are limited to inquiring about what the worker thinks
the employer knew when the worker was hired.
4 Survey Design
Our survey is part of the Princeton Data Improvement Initiative, a project to develop new
questions for labor force surveys. The questionnaire was patterned after the Current Popula-
tion Survey and included questions on career experience, job tasks, and occupational licens-
ing. We designed a module to assess the prevalence of wage posting and employer-employee
bargaining at the time employees were hired. Based on a focus group, we concluded that
individuals who were hired within the past 10 years could recall how knowledgeable they
were about the pay on their job when they first interviewed for it, whether the employer
made a take-it-or-leave-it offer, whether they could have remained on their previous job if
they had wanted to, and whether their employer was aware of their pay on their previous
job prior to making them an offer. Those who were employed at the time of the survey were
asked about their current job (87 percent), and those who were unemployed at the time of
the survey were asked about their last jobs (13 percent). The unemployment rate for the
(weighted) sample as a whole was 5.3 percent.
The survey organization Westat conducted the survey from June 5 to July 20, 2008.
Individuals age 18 or older who were in the labor force were eligible for the survey. A
total of 2,513 individuals were interviewed, 1,435 of whom were hired in the previous 10
years. Westat used a random-digit-dial sampling design constructed from a national sampling
frame of residential exchanges. The selected numbers were then called and screened to
identify households with eligible respondents. One respondent was randomly selected from
each eligible household for the interview using the nearest birthday procedure. Up to 15
callbacks were made to try to elicit responses. Some 28 percent of sampled eligible households
17
agreed to participate in the screening questions, and 64 percent of the selected individuals in
screened households completed the questionnaire. Thus the response rate was 17.9 percent,
using the American Association for Public Opinion Research response rate definition 3 (see
aapor.org/uploads/Standard Definitions 04 08 Final.pdf, p. 35).
Westat developed survey weights to compensate for variation in selection probabilities,
differential response rates, and possible under-coverage of the sampling frame. The derivation
of the sample weights focused primarily on matching the marginal distributions of the Cur-
rent Population Survey by sex, age, educational attainment, census region, urbanization,
race, Hispanic ethnicity, employment status, and class of employer (private, government,
etc.). See irs.princeton.edu/PDIIMAIN.htm for a detailed description of the derivation of
the sample weights and the questionnaire.
Although the survey response rate is low compared to many government labor force
surveys, it is comparable to that in commercial surveys. Groves and Peytcheva (2008) show
that survey non-response rates by themselves are not associated with significant bias. Low
response rates are a concern when the causes of participation in the survey are correlated
with the survey variables of interest. We do not believe that wage-formation practices from
years earlier would be correlated with survey participation. The response rate was low in
large part because many households declined to participate in the screener questions, which
did not mention wages or job search at all. Another reason for placing some confidence in
the representativeness of our sample is that a standard Mincerian wage regression using data
from the survey closely matched the corresponding regression from the Current Population
Survey. Although we would have preferred a higher response rate, we have no reason to
believe that non-response skews our results in favor or against any particular wage formation
model
5 Findings
5.1 Descriptive logit model
To describe our survey findings, we use a logit probability model for yes-no variables con-
structed from the respondents’ answers. The model predicts the probability of a yes answer,
given a set of variables describing the individual and the job. These variables are
• Indicator for African-American individual
18
• Indicator for Latino or Latina individual
• Indicator for a woman
• A set of indicators for education, in five categories
• Indicator for union membership
• A set of indicators for private, government, and non-profit employer
• Work experience in years
• Age in years
• Tenure in years
• Indicator for repetitive job
• Indicator for physical job
• Indicator for job involving managing
• Indicator for job involving problem solving
• Indicator for job involving use of math
• Indicator for job involving reading long documents frequently
Table 1 summarizes the composition of the survey respondents.
We use a weighted logit estimator because the purpose of estimation is to describe the
responses, not estimate underlying parameters. We use the resulting logit model to make
statements about responses in different subsets of the population. We present the results in
terms of the estimated probability of a yes answer for a variety of types of workers, along with
bootstrap standard errors of the probabilities and of the differences between the probability
for a group and the probability for a base case. The online backup materials for this paper
include the underlying logit estimates.
Our base case is: individual not African-American, not Latino or Latina, a man, high-
school education but no college, not a union member, working for a private employer, 40
years old, 20 years of experience, 4 years of tenure, and none of the specific job characteristics
listed above. We display the results as probabilities of a yes answer for a variety of groups
19
Category or characteristic Percent or average
African-American 10
Latino/a 15
Woman 46
Education
Not HS graduate 9
Some college 25
College graduate 22
Professional training 13
Union member 17
Government job 17
Non-profit job 11
Years of work experience 18
Age 41
Currently employed 88
Lost job in past 3 years 13
Years with this employer 7
Repetitive activities 50
Physical activity 57
Managing or supervising 29
Solving problems 71
Use of advanced math 26
Reading long documents 17
Table 1: Survey Respondents
20
defined by the right-hand variables. In addition to groups defined by a single indicator,
such as for women, we include four groups defined by combinations of right-hand variables
(variables not mentioned in this list are the same as in the base case):
• Senior: 40 years of experience, 60 years old, 30 years tenure, job involves managing
• Knowledge worker: post-college education, 20 years of experience, 60 years old, 30
years tenure, job involves solving problems, using advanced math, and reading long
documents
• Blue collar: union member, 20 years of experience, 40 years old, 10 years tenure, job
involves physical and repetitive tasks
• Recent job loser: 20 years of experience, 40 years old, one year of tenure, job involves
none of the specific characteristics
5.2 Evidence about the relative importance of wage posting andbargaining
Table 2 summarizes the responses to the question, “When you were offered your (cur-
rent/previous job), did your employer make a ‘take-it-or leave-it’ offer or was there some
bargaining that took place over the pay?” The table describes the probability that a respon-
dent would answer that some bargaining occurred. A respondent with the base character-
istics has a probability of 31 percent of that response. As we discussed earlier, the absence
of bargaining implied by that response does not necessarily mean that the respondent be-
lieved that the employer was committed to the wage offer and would not have entertained a
counteroffer. In an alternating-offer equilibrium, the job-seeker can make a counteroffer and
the employer would consider it, but the job-seeker never does, because the original offer was
made with that possibility in mind and was high enough to make a counteroffer not worth
the effort. Thus, in some ways, it is a surprise that 31 percent in the base group replied that
some bargaining did take place. The observed incidence of bargaining arises from departures
from the assumptions of the full-information alternating-offer bargaining game. The depar-
tures could include private information, potentially on both sides, and biased assessments of
worker and job characteristics.
The frequency of no-bargaining responses varies substantially among job-seekers. It is
higher than the base-case level among African-Americans (43 percent) and Hispanics (44
21
Probability of bargaining,
percent
Difference from base
case
Base case 31(7)
African-American 43 11(11) (9)
Latino/a 44 12(10) (7)
Woman 24 -8(6) (4)
Not HS graduate 28 -3(10) (9)
Some college 40 9(8) (6)
College graduate 42 11(7) (6)
Professional training 56 25(7) (7)
Union member 14 -18(5) (5)
Government job 16 -15(6) (5)
Non-profit job 25 -7(7) (4)
Senior 43 12(18) (16)
Knowledge worker 86 55(4) (7)
Blue collar 5 -26(2) (7)
Recent job loser 32 0(7) (2)
Note: Bootstrap standard errors in parentheses.
Based on response, some bargaining took place
Table 2: Probability that Some Bargaining Occurred over Pay
22
percent). Women, at 24 percent, are rather less likely than the men in the base case to bar-
gain. The incidence of wage bargaining rises dramatically with education. Respondents with
professional education had a probability of 56 percent of a bargaining during hiring. Finally,
and not surprisingly, union members (14 percent) and government workers (16 percent) had
low propensities to report bargaining over pay. Our other cases show dramatic variation in
the incidence of bargaining. Knowledge workers, at 86 percent, almost all reported bargain-
ing, whereas blue-collar workers, at 5 percent, almost never bargain. Senior workers, at 43
percent, are in the middle.
Table 3 describes the answers to the question, “At the time that you were first interviewed
for your job, did you already know exactly how much it would pay, have a pretty good idea
of how much it would pay, or have very little idea of how much it would pay if you got it?”
We consider the probability of the answer that the respondent knew exactly how much it
would pay. We believe that this answer would be chosen by individuals who had applied for
a job after seeing a formal description that included a committed rate of pay. We do not
show the results for the group who responded that they knew exactly or had a pretty good
idea, because the responses for all groups were high—uniformly above 80 percent. It would
have been useful for the survey to include another response somewhere between “exactly”
and “pretty good idea.”
In the base case, 22 percent of the respondents in the base group reported that they
knew exactly how much the job paid before the employer learned about the respondent. The
weighted unconditional sample mean is 31 percent. The difference arises from the fact that
the base group is not representative of the entire sample.
On its face, this evidence suggests that nearly a third of jobs involve posted wages. We
do not push this interpretation too far, because, on the one hand, another 65 percent of the
respondents in the base group said they had a pretty good idea of what the job would pay,
and, on the other hand, job-seekers could know the wage even if it were not a committed,
posted wage. In standard bargaining models with no private information, the worker knows
in advance what wage will result from bargaining.
The table shows that an African-American worker otherwise in the base group has a
somewhat lower likelihood, 21 percent, of knowing the pay in advance, while a Latino or
Latina has an even lower likelihood, 17 percent. Women have the same likelihood as men.
The probability of knowing pay in advance falls substantially with education. Union members
23
Probability, percent
Difference from base
case
Base case 22(5)
African-American 21 -2(8) (6)
Latino/a 17 -6(6) (5)
Woman 22 0(5) (3)
Not HS graduate 20 -2(7) (6)
Some college 20 -3(5) (4)
College graduate 15 -8(4) (4)
Professional training 14 -8(4) (5)
Union member 36 13(8) (6)
Government job 36 14(8) (6)
Non-profit job 27 4(7) (4)
Senior 53 30(17) (16)
Knowledge worker 17 -6(4) (6)
Blue collar 56 33(8) (8)
Recent job loser 20 -2(5) (1)
Note: Bootstrap standard errors in parentheses.
Worker knew pay exactly prior to interview
Table 3: Probability of Knowing What the Job Would Pay
24
and those who took government jobs report knowing the wage exactly with substantially
higher frequency. Except for a reversal of the pattern of education effects, the differences
among respondents captured by the variables are similar. Apparently the most educated
workers are the least likely to admit they were clueless about pay when they applied for a
job.
Do employers determine and post wages prior to screening workers or do they make an
offer to a worker after screening that is, in principle, negotiable via a counteroffer? No
single question in the survey answers this important question. The results above showed
that about a third of workers know wages exactly prior to their interviews and that about
two-thirds viewed their pay offer as having a take-it-or-leave-it character. The left panel
of Table 4 describes the respondents who said they knew the pay exactly prior to being
interviewed and that there was no bargaining over pay. The likelihood that a base-case
respondent gave these two answers is 15 percent. As an estimate of the fraction of workers
whose wages were posted, this estimate has biases in both directions. It is an underestimate
if workers felt they did not know the pay exactly, even though they were well informed. It
is an overestimate on account of workers who anticipated how bargaining would later come
out, but received a customized wage influenced by the employer’s inability to disregard a job-
seeker’s counteroffer. The essence of the posted-wage model is the employer’s commitment
to disregard counteroffers.
Table 4 shows large variations across categories of workers in the estimated incidence of
wage posting based on the criterion of knowing the wage in advance and not engaging in
bargaining. African-Americans and Hispanics face slightly lower likelihoods, at 12 percent
and 10 percent. Women are higher than the base value, at 17 percent. The incidence of
wage posting declines dramatically with education, from 12 percent for those who did not
complete high school to 5 percent for those with professional training. The higher incidence
of wage posting for the least educated is consistent with the view that a wage constrained by
the minimum wage is inherently posted. Somewhat more than 10 percent of the respondents
earned the minimum wage.
At 28 percent, wage posting is far more common for union members. Similarly, govern-
ment jobs, at 31 percent, are substantially more likely to have posted pay, compared to the
base case. The logit coefficients in the right-hand panel are generally similar to those in the
left-hand panel. Most of the difference comes from the much higher constant for the more
25
Probability, percent
Difference from base
case
Base case 15(4)
African-American 12 -3(6) (4)
Latino/a 10 -5(4) (4)
Woman 17 3(5) (3)
Not HS graduate 12 -3(6) (6)
Some college 12 -3(4) (3)
College graduate 9 -6(3) (3)
Professional training 5 -9(2) (4)
Union member 28 13(7) (5)
Government job 31 16(9) (6)
Non-profit job 22 7(7) (5)
Senior 27 13(17) (16)
Knowledge worker 4 -10(1) (4)
Blue collar 50 36(9) (10)
Recent job loser 13 -2(4) (1)
Bootstrap standard errors in parentheses.
Response: knew wage exactly and wage was take-it-or-leave-
it
Table 4: Probability of Wage Posting
26
inclusive criterion, so the probabilities on the right are roughly proportionally higher than
those on the left.
Table 4 indicates a higher incidence of wage posting in the more standardized jobs avail-
able to those who have not graduated from college and the lower incidence among college
graduates and those with professional education. The highest probability of posting in the
table is 50 percent for blue collar workers and the lowest is 4 percent for knowledge workers.
5.3 Evidence about factors that influence bargaining
Table 5 summarizes the responses to the question, “Think back to the time when you were
offered your (current/most recent) job. When you were offered this job, was it possible for
you to keep your previous job instead if you wanted to?” Our interest in this topic derives
from the value of the option to keep a current job in a bargaining setting. The sample
includes those who were not employed immediately prior to obtaining their most recent job
(coded as unable to keep their previous job). The left panel describes the answers among all
respondents and the right panel reports the frequency of bargaining over pay among those
who could have kept the previous job at the time they accepted the current job.
Among all respondents, in the left-hand panel, the table shows that an individual in the
base category had a 48 percent chance of answering yes. Thus almost half of job-seekers
has the bargaining advantage of the option of keeping an existing job. Variations from the
base-case probability of retaining a previous job are relatively small, according to the table.
Minority members are slightly more likely to retain the option and women slightly less likely.
The likelihood of the option is a bit lower for the least educated and a bit higher for college
graduates, though just the same as in the base case for those with graduate training. Union
members are also slightly more likely to have the option of keeping an existing job. Note
that the fractions of job-seekers with the option are necessarily higher than the figure in
the table—our data omit instances in which employed job-seekers decided that a new job
was not as desirable as their existing job and therefore remained at the job despite finding
another employment opportunity. Our survey focused on the beginning of the current or
most recent job and did not inquire about job offers received in the course of that job.
The incidence of actual bargaining among those who could have kept their previous
jobs varies tremendously. In the base case, 45 percent bargained, rather higher than the
31 percent in Table 2 for all workers. Among workers in the senior group who could have
27
Probability, percent
Difference from base
case
Probability, percent
Difference from base
case
Base case 48 45(7) (13)
African-American 54 7 66 22(10) (8) (14) (11)
Latino/a 50 3 69 25(9) (6) (16) (10)
Woman 46 -1 37 -8(7) (4) (13) (8)
Not HS graduate 49 1 38 -7(10) (9) (15) (16)
Some college 50 2 62 17(7) (6) (13) (10)
College graduate 48 1 64 20(6) (6) (11) (10)
Professional training 40 -7 83 38(7) (6) (8) (11)
Union member 56 8 10 -35(9) (6) (6) (9)
Government job 43 -5 27 -18(9) (6) (12) (10)
Non-profit job 46 -2 48 3(9) (7) (16) (10)
Senior 42 -6 78 33(16) (15) (21) (19)
Knowledge worker 48 0 90 45(8) (9) (4) (13)
Blue collar 48 0 3 -42(9) (9) (2) (12)
Recent job loser 49 1 44 -1(8) (2) (13) (3)
Note: Bootstrap standard errors in parentheses.
Could have kept earlier jobAmong those who could have
kept earlier job, some bargaining occurred
Table 5: Probability that the Previous Job Could Have Been Kept
28
kept their jobs, 78 bargained, also well above their bargaining propensity among all workers.
Thus the role of the option to keep the current job when considering a new job opportunity
in influencing the wage through bargaining is substantial, especially among more educated,
problem-solving workers. The recent attention to the economics of on-the-job search is fully
merited, in light of these findings.
Job-ladder models, such as Hagedorn and Manovskii (2009), stress the importance of
the wage at one job that can be retained in negotiating with a prospective new employer.
The models typically assume that the option to keep the earlier job was available for any
job-to-job transition that is observed without intervening unemployment. With our data,
we can take a different approach by asking if the current wage is higher among those who
retained the earlier option against those who did not. The first set of workers are at least one
step up the job ladder whereas the second set must be at the bottom. To investigate this, we
estimate an equation for the median wage (using quantile estimation) with all of the variables
in our descriptive models except the job characteristics, plus a dummy variable for those who
answered that they could have retained their earlier jobs. We found that those with that
advantage had a wage $2.07 per hour higher than otherwise similar workers without the
advantage, with a bootstrap standard error of $0.71. The evidence unambiguously supports
the hypothesis of a job ladder.
Table 6 describes the answer to the question, “Did your [current/most recent] employer
learn how much you were making in your previous job before making you your job offer?”
Knowledge of earlier pay is useful to the employer in cases where the possibility of bargaining
influences the wage. The likelihood of a yes answer is 48 percent in the base case. Respon-
dents with other characteristics varied only a small amount from this value. As expected,
employers learned earlier pay less frequently for union members and for government jobs,
but the difference is small.
The finding that many employers made an effort to learn earlier pay rates gives some
further support to the hypothesis that wage posting is not the dominant mode of wage
formation.
6 Evidence on Bargaining and Heterogeneity
As we discussed earlier, dispersion should be higher among workers who bargained relative
to those who accepted a posted wage, within a category of workers defined by observed and
29
Probability, percent
Difference from base
case
Base case 48(8)
African-American 37 -11(10) (7)
Latino/a 47 -1(10) (6)
Woman 52 5(7) (4)
Not HS graduate 42 -5(11) (9)
Some college 41 -7(7) (5)
College graduate 44 -3(8) (6)
Professional training 49 1(8) (7)
Union member 43 -5(9) (6)
Government job 43 -5(9) (6)
Non-profit job 48 0(9) (5)
Senior 50 3(16) (15)
Knowledge worker 52 5(8) (10)
Blue collar 42 -6(8) (9)
Recent job loser 48 0(8) (2)
Note: Bootstrap standard errors in parentheses.
Employer knew previous pay
Table 6: Probability that the Employer Learned Previous Pay before Making Job Offer
30
verifiable characteristics. Theory suggests that wages will be higher for workers who bargain.
We examine these predictions in this section.
We consider the distribution of the wage, w, conditional on a dummy variable b = 1 for
a worker who bargained and 0 if not. This distribution varies by observed characteristics
x. We write the distribution as Fb(w;x), so F0(w;x) is the distribution of w among non-
bargainers and F1(w;x) the distribution among bargainers. We expect that the dispersion of
F1(w;x) will exceed that of F0(w;x)—wage dispersion will be greater among the bargainers
than the non-bargainers.
We measure the distributions Fb(w;x) by quantile estimation. This method finds values
wi,b(x) such that Fb(wi,b(x)) = qi for a set of probabilities qi. We take these to be deciles, so
qi = i/10 for i = 1, . . . , 9. If the theoretical proposition is correct that dispersion is greater
with bargaining, then dispersion measures such as w9,b−w1,b, the spread between the 9th and
1st deciles, should be greater among the bargainers, b = 1, than among the non-bargainers.
See Koenker and Hallock (2001) for a discussion of quantile estimation.
We take the quantile functions to be linear in the characteristics:
wi,b(x) = γi,bx. (12)
Of course, we choose the variables x to be nonlinear functions, such as dummy variables for
categories, of the actual measures from our survey.
Figure 2 compares the distribution of wages among non-bargainers, on the left, and
bargainers, on the right. These are the overall marginal distributions over the characteristics
x, computed directly from the wage data without quantile estimation. The shading in
the vertical bars designates the deciles of the distribution, from the first to the ninth. The
dispersion of wages is substantially higher among the bargainers, in line with the theory. And
the median of wages is higher among bargainers, again in line with theory. The evidence
from the raw dispersion of wages is far from dispositive, however, because the theory deals
with dispersion conditional on the observed characteristics of job applicants. Education is
the most salient of those characteristics and unquestionably an observed characteristic that
plays an important role in screening workers.
Figure 3 repeats the comparison of the distribution of wages, now in a setting that stan-
dardizes for demographics and education, using quantile estimation. It shows the distribution
of the wage of a white male with a high-school education but no college. We fit the distri-
bution to all of our data, using quantile estimation at 9 deciles, then calculate the implied
31
6
7
8
9
9
30
40
50
60
70
ourly wage, dollars
1 12
23
34
4
5
5
67
8
0
10
20
Non‐bargainers Bargainers
Ho
Figure 2: Distributions of Wages among Non-Bargainers and Bargainers
wages at the deciles, as displayed in the figure. Because bargaining is more common among
better-educated workers, wage heterogeneity is smaller after conditioning on education. The
dispersion remains higher among bargainers and so is the difference between median wages
in the two groups.
In addition to the specification illustrated in Figure 3, we also consider a more restricted
specification in which the characteristics x determine the medians of the two distributions in
an unrestricted way, but that the only difference in dispersion around the median is related to
bargaining and not to the x-variables. To measure the difference in wage dispersion between
the two groups for that assumption, we find the quantiles of the wage centered on the median;
that is, we tabulate the deciles of the variables wi−mbi(xi), where wi is worker i’s wage and
mbi(xi) is the median estimated for the worker’s characteristics xi and bargaining dummy
bi.
Table 7 shows the results of these calculations. The left column shows the difference in
the inter-decile range between bargainers and non-bargainers and the right column shows
the differences in the median wage. In all cases we show the bootstrap standard errors of
the differences. The top line examines the raw data on hourly earnings. The inter-decile
range is $23 per hour greater among bargainers than among non-bargainers and the median
32
9
9
8
8
5
6
7
7
20
25
30
35
40
45
ourly wage, dollars
1 1
223
34
4
5
6
0
5
10
15
Non‐bargainers Bargainers
Ho
Figure 3: Distributions of Wages among Non-Bargainers and Bargainers, for White MaleHigh-School Graduates
wage is $11 per hour greater. The results confirm that the dispersion among bargainers is
substantially higher and that this finding cannot be the result of sampling error alone.
The second line of the table shows the results of the more restricted specification in which
demographics and education (the African-American, Latino/a, sex, and education dummy
variables) affect the median wage but not the dispersion around the median. The difference
in the dispersion of the raw wage around the fitted median between bargainers and non-
bargainers is a little under $17 per hour, with a bootstrap standard error of just over $3 per
hour (the bootstrap procedure takes full account of the two step estimation). Again, the
evidence in favor of higher dispersion among bargainers is statistically unambiguous. The
difference between the median hourly wages of bargainers and non-bargainers falls to $8 with
the adjustment, but the difference remains substantial in economic terms and statistically
unambiguous. We conclude that there is strong evidence that workers who bargain earn
higher wages, after considering that higher education raises both the level of wages and the
likelihood of bargaining. Of course, the conclusion rests on the hypothesis that the demo-
graphic and education variables account for all the differences in wages that are correlated
with bargaining. We are inclined to believe that part of the large advantage found here for
33
Dispersion: Difference between 9th and 1st decile wage, dollars per hour
Level: Median, dollars per hour
Without adjustment for demographics and education 23.00 11.00(3.50) (1.20)
With adjustment for demographics and education in median wage but not in dispersion, measured for white male high-school graduates 16.67 8.17
(3.26) (2.29)
With adjustment for demographics and education in median and disperson, measured for white male high-school graduates 9.68
(6.26)
Bootstrap standard errors in parentheses
Measure of difference between bargainers and non-bargainers
Adjustments for demographics and education
Table 7: Differences in Wage Distributions between Bargainers and Non-Bargainers, forWhite Male High-School Graduates
bargainers is the result of the factors discussed earlier in the paper that point in the direction
of an advantage, but recognize that part of the large difference we find may be the result of
selection factors unrelated to those factors.
The bottom line of Table 7 shows the difference in the inter-decile range for our base
case when the entire distribution is taken to depend on the demographic and education
variables, not just the central tendency. The difference in the inter-decile range is now
somewhat below $10 and its standard error is above $6. Contrary to the assumption in
the middle line of the table, some of the difference in dispersion between bargainers and
non-bargainers disappears when the measure is standardized for the differing compositions
of the two groups as measured by the demographic and education variables. The evidence
remains favorable to the hypothesis that wage heterogeneity is higher among bargainers,
but sampling variation is substantially higher and the measured effect is smaller. The large
increase in the standard error in moving from the second line of the table to the third arises
from the quantile estimation of the way that the characteristics x influence the dispersion.
We intend our examination of wage dispersion mainly to validate the survey’s measure
of bargaining. Because posted wages inherently have less dispersion than bargained wages,
no matter what governs employers’ choices to post wages, we cannot claim to have validated
a theory of when posting occurs. Apart from the characteristics that we and presumably
34
employers observe, we do not measure the remaining differences in skills among workers who
appear identical in terms of those characteristics, whereas the theory of the choice between
posting and bargaining considers exactly those differences.
7 Conclusions
The two leading models of individual wage formation are both important in the labor market
of the United States. Between a quarter and a half of workers hold jobs that were filled with
posted wages. Jobs held by women and by people with little education are more likely to
have been posted-wage positions. College graduates and those with professional training
are rather unlikely to hold posted-wage jobs. Posted wage jobs are also common in the
government and union sector.
The evidence suggests that most of the remaining workers do not face posted wages but
could make a counteroffer that an employer could not resist considering. About a third of
workers report explicit bargaining over pay. From bargaining theory, it is reasonable to infer
that a fair number of others could have made a counteroffer, but employers, recognizing that
possibility, make a satisfactory initial offer.
Wage-formation theory emphasizes the importance of the option that a job-seeker may
have to retain a current job. The option is powerful in forcing a prospective employer to bid
high to hire the worker. We find that 48 percent of workers in the base group had this option
when they took their current jobs. On-the-job search is a central feature of the U.S. labor
market. Job-ladder models emphasizing the bargaining power of a worker who already has a
high-paying job in negotiating with a prospective employer are on the right track, especially
among knowledge workers.
We find that about half of workers report that their new employers learned the workers’
earlier pay rates before making them job offers. Employer interest in earlier pay is an
indication against wage posting. Employers presumably use the information to formulate a
satisfactory offer to workers who retain the option on earlier jobs. The information is not
sufficiently widespread and employers are unable to commit to ignore counteroffers, else the
U.S. labor market would suffer from the Diamond paradox. Instead, many workers appear
to earn substantial Ricardian rents.
35
References
Ausubel, Lawrence M., Peter Cramton, and Raymond J. Deneckere, Handbook of Game
Theory, Elsevier, 2002.
Binmore, Ken, Ariel Rubinstein, and Asher Wolinsky, “The Nash Bargaining Solution in
Economic Modeling,” RAND Journal of Economics, Summer 1986, 17 (2), pp. 176–188.
Burdett, Kenneth and Dale T. Mortensen, “Wage Differentials, Employer Size,and Unem-
ployment,” International Economic Review, May 1998, 39 (2), 257–273.
and Kenneth L. Judd, “Price Dispersion,” Econometrica, July 1983, 51 (4), 955–969.
Butters, Gerard R., “Equilibrium Distributions of Sales and Advertising Prices,” Review of
Economic Studies, October 1977, 44 (3), 465–491.
Christiano, Lawrence J., Martin Eichenbaum, and Charles L. Evans, “Nominal Rigidities
and the Dynamic Effects of a Shock to Monetary Policy,” Journal of Political Economy,
2005, 113 (1), 1–45.
Diamond, Peter A., “A Model of Price Adjustment,” Journal of Economic Theory, 1971, 3,
156–168.
, “Wage Determination and Efficiency in Search Equilibrium,” Review of Economic
Studies, April 1982, 49 (2), 217–227.
and Eric Maskin, “An Equilibrium Analysis of Search and Breach of Contract, I:
Steady States,” Bell Journal of Economics, 1979, 10, 282–316.
Ellingsen, Tore and Asa Rosen, “Fixed or Flexible? Wage-Setting in Search Equilibrium,”
Economica, 2003, 70, 233–250.
Groves, Robert M. and Emilia Peytcheva, “The Impact of Nonresponse Rates on Nonre-
sponse Bias: A Meta-Analysis,” Public Opinion Quarterly, 2008, 72 (2), 167–189.
Hagedorn, Marcus and Iourii Manovskii, “Spot Wages over the Business Cycle?,” October
2009. Department of Economics.
Hall, Robert E., “Employment Fluctuations with Equilibrium Wage Stickiness,” American
Economic Review, March 2005, 95 (1), pp. 50–65.
36
and Edward P. Lazear, “The Excess Sensitivity of Layoffs and Quits to Demand,”
Journal of Labor Economics, 1984, 2 (2), 233–257.
and Paul R. Milgrom, “The Limited Influence of Unemployment on the Wage Bargain,”
American Economic Review, September 2008.
Koenker, Roger and Kevin F. Hallock, “Quantile Regression,” Journal of Economic Per-
spectives, Autumn 2001, 15 (4), 143–156.
McCall, John J., “Economics of Information and Job Search,” Quarterly Journal of Eco-
nomics, February 1970, 84 (1), 113–126.
Menzio, Guido, “A Theory of Partially Directed Search,” Journal of Political Economy,
2007, 115 (5), 748–769.
Michelacci, Claudio and Javier Suarez, “Incomplete Wage Posting,” Journal of Political
Economy, 2006, 114 (6), 1098–1123.
Mortensen, Dale T., “Specific Capital and Labor Turnover,” Bell Journal of Economics,
Autumn 1978, 9 (2), 572–586.
, Wage Dispersion: Why Are Similar Workers Paid Differently?, Cambridge, Mas-
sachusetts: MIT Press: Zeuthen Lecture Book Series, 2003.
and Christopher Pissarides, “Job Creation and Job Destruction in the Theory of
Unemployment,” Review of Economic Studies, 1994, 61 (0), pp. 397–415.
Moscarini, Giuseppe and Fabien Postel-Vinay, “The Timing of Labor Market Expansions:
New Facts and a New Hypothesis,” NBER Macroeconomics Annual, 2008, pp. 1–51.
Pissarides, Christoper A., Equilibrium Unemployment Therory, second ed., Cambridge,
Massachusetts: MIT Press, 2000.
Rogerson, Richard and Robert Shimer, “Search in Macroeconomic Models of the Labor
Market,” in Orley Ashenfelter and David Card, eds., Handbook of Labor Economics,
Vol. 4, Elsevier, 2012, forthcoming.
, , and Randall Wright, “Search-Theoretic Models of the Labor Market: A Survey,”
Journal of Economic Literature, December 2005, 43, pp. 959–988.
37
Shimer, Robert, Labor Markets and Business Cycles, Princeton, NJ: , Princeton University
Press, 2010.
Stigler, George J., “The Economics of Information,” Journal of Political Economy, 1961, 69
(3), 213–225.
, “Information in the Labor Market,” Journal of Political Economy, 1962, 70 (5, Part
2), 94–105.
38
top related